Rambus, Pentium 4: The Hall of Mirrors

Back in June we revealed that Taiwanese OEMs were holding off implementing the first rev of the Pentium 4, because only two or three quarters after introduction, which we believe will happen in September, the Tulloch chipset supporting a 479 pin version of the microprocessor will supersede it.

This week's admission by Intel that it is developing a PC-133 solution for the Pentium 4, probably ready in the second half of next year, only serves to remind us of that old fairground attraction known as the Hall of Mirrors, with everything distorted in whichever direction you look.

This news, which was actually revealed by German magazine PC Welt back in May, has caused the Rambus share price to behave like a Bucking Bronco again. Yesterday, its share price closed at $66.6875. Since the stock split, its high was $127 on June 23.

At the Intel Developer Forum last February, we repeatedly asked Peter MacWilliams, a top Intel executive and an Intel Fellow, no less, whether there was any possibility whatever that his company would introduce double data rate (DDR) memory for the Willamette.

Earlier that day, Dr Albert Yu, a senior Intel VP, had demonstrated the Willamette to a packed press room, filled to overflowing not just with hacks but with a good sprinkling of Wall Street financial analysts. As his presentation continued, and as we tapped our story into a PC, we were more than a little surprised to hear two financial analysts sat bang next to us buying and selling Rambus stock, as Yu can imagine. The Rambus share price soared to dizzy heights during the course of that day, after a period where it languished in one of its troughs of despond.

We are now asked to believe that Intel has taken a fresh view, and over the last couple of weeks, that PC-133 is, for some reason or other, the right kind of memory for the Pentium 4. MacWilliams then found it hard to answer the question why Willamette would use Rambus memory and Foster, the server version of the same microprocessor, would use DDR memory.

Why, we might instead ask, has the Pentium 4 mirror now become concave rather than convex, and Intel's premier IA-32 chip will eventually use PC-133 memory, inferior to DDR memory and, if MacWilliams and a gaggle of other Intel bodies are to be believed, vastly inferior to Rambus?

And, for this matter, does this latest excursion onto the Memory Bucking Bronco mean that Intel has now torn up its existing roadmaps and wants to start from scratch again?

We're tempted to say "who bloody knows?" but will, instead, attempt to have a stab at understanding this latest szechuan.

Intel will have three quarters or so to flog the Pentium 4 in its various speed revs, and using Ramboid RIMMs, before 479-pin Willamette and the Tulloch chipset emerges from the deep. (That's assuming Intel hasn't torn up the 479-pin version).

Just one quarter or so after that, according to the latest info, we will begin to see PC-133 versions of a chipset for the Pentium 4, perhaps.

In between times, Intel will introduce the Foster microprocessor. Although that was, according to MacWilliams and other, supposed to be using a DDR chipset, roadmaps we saw just a few weeks ago seemed to suggest that Rambus could now be an option for Foster too.

How is Intel's i815 chipset doing? This chipset, pre-launched at this year's Computex, represented Intel's partial climbdown over PC-133 memory caused, mainly, because PC vendors and Taiwanese mobo makers were unhappy with Caminogate (the i820).

We are now given to understand that the big US vendors, including Dell and Gateway, have moved to i815 based chipsets in many of their systems.

And where are the DDR chipsets? We searched in vain at Computex for a sign of the times. ALi was "demoing" its DDR chipset, but back in June it wasn't working, and, realistically, it will be October or so before we see any of these in the real world.

Intel, Rambus and Kingston Technology have told us during the course of this year that the price of RIMMs is falling and will fall. Yesterday, there was a further price cut in the price of Rambus. (Check out this informative page for pricing). The price of SDRAM has risen.

So, perhaps, Intel has a cunning plan to take advantage of these movements, and flog as many Pentium 4-Tehama systems as it can, starting in September and continuing through the next three quarters. When we managed to get hold of the projected pricing of the Pentium 4, we were surprised at how inexpensive it was. And maybe all of the above is part of its reasoning. By the time its PC-133 Pentium 4 chipset is out, this time next year, it will have migrated the price of this microprocessor downwards, and introduced several speed and big cache revs based on Willamette technology.

Being as paranoid about Via as it admits it is, it will be hoping that by this time next year, it has an offering for as many levels in the marketplace as possible. It's just that 479 pin version of Willamette that's bugging us a little...

Sorry, there's going to be a lot of links to look at below. ®

See Also

The Dramurai
Intel does u-turn on Willamette and synchronous RAM (May this year)
Pentium 4 to use SDRAM
Cheap Rambus memory on way
AMD loose cannon in Intel-Via deal
Via gives DDR ringing endorsement
Rambus rambushes DDR camp
Where the hell... are DDR chipsets?
Rambus exec kicked out of DDR seminar
Mosel predicts rapid death of SDRAM

Roadmaps

Computex 2000 coverage
Willamette to have triflingly short life
Intel server maps show Foster incursion
Intel's cunning server plans
Intel revamps boxed chip roadmaps
Tulloch, Willamette plans firm up
Intel's view on DDR
Intel roadmap times, they are a changin...
Pentium 4 pricing revealed
Intel's pricing: May to September

Chip Pornography
Pentium 4 pictures revealed